A Novel Method for Disentangling Strategies from Visual Search
Vicente Pallar
´
es
1
, Lorena Rami
2
and Laura Dempere-Marco
1,3
1
Department of Information and Communication Technologies, Universitat Pompeu Fabra, Barcelona, Spain
2
Alzheimer’s Disease and Other Cognitive Disorders Unit, IDIBAPS, Hospital Clinic, Barcelona, Spain
3
Faculty of Sciences and Technology, Universitat de Vic-Universitat Central de Catalunya, Vic, Spain
Keywords:
Eye-Tracking, Feature Domain Analysis, Visual Search, Aging.
Abstract:
The process of actively scanning a visual scene while looking for something in a cluttered environment is
known as visual search. In this work, we show that it is possible to disentangle the strategies pursued by
subjects to solve visual tasks by investigating dynamical aspects inherent to eye-tracking data. A novel method
is proposed to characterize visual search strategies in a generalized N-dimensional feature domain, which
allows us to investigate spatial-temporal aspects of the search as well as the subjects’ reliance on visual cues.
In order to validate the proposed method, we have developed an experimental paradigm based on a double
conjunction search in which one the visual cues is systematically manipulated, which can induce feature-
based strategies in the observers. On the basis of the preliminary evidence presented here, we argue that this
characterization of visual search strategies opens new avenues to assess cognitive function and its relation to
normal aging.
1 INTRODUCTION
It is well-established that we do not systematically
scan the visual world with our eyes when looking for
something in a cluttered environment. Instead, we al-
locate the gaze selectively so as to maximize the in-
formation that is relevant for solving the task at hand.
Thus, much information can be retrieved from the
way the brain processes visual information by study-
ing the eye movements during the exploration of a
particular scene. For this reason, visual exploration
mechanisms have been investigated for many years,
and from many different perspectives, e.g. psypho-
physics (Treisman and Gelade, 1980; Wolfe, 1998),
computational neuroscience (Deco and Rolls, 2004;
Deco and Zihl, 2006), or statistical physics (Boc-
cignone and Ferraro, 2004). One important technique
that tries to exploit these mechanisms and has become
more used during the last decades is eye-tracking.
Several cognitive processes have been, and are being,
studied by means of eye movements, like scene per-
ception (V
˜
o and Henderson, 2010), visual search (see
(Yang et al., 2002; Dempere-Marco et al., 2011) for a
review) or reading (Rayner, 2009).
Arguably, visual attention plays a critical role in
strategy formation and maintenance. Interestingly,
a close relation between eye movements and visual
attention has been broadly accepted. In particular,
two main mechanisms to reallocate attention during
visual search have been identified. The first one,
called overt attention, relies on shifting the gaze to-
ward a new location, whereas the second one, known
as covert attention, consists of paying attention to
an area in the periphery of the foveal region but
without redirecting the gaze. Evidence of the exis-
tence of both types of attention has been proven by
means of psychophysics, electrophysiology and neu-
roimaging (Carrasco, 2011). The interaction between
saccade programming and covert attention has also
been confirmed (Deubel and Schneider, 1996; Hoff-
man and Subramaniam, 1995), even unveiling the
existence of common neural mechanisms (Corbetta
et al., 1998; Beauchamp et al., 2001; Grosbras et al.,
2005). A similar link between fixational eye move-
ments (especially microsaccades) and covert atten-
tion has been largely suggested (Engbert and Kliegl,
2003; Laubrock et al., 2010; Yuval-Greenberg et al.,
2014). Of note, the visual search paradigm has been
key in attention research for investigating visual at-
tention deployment, both overt and covert (e.g. Deco
and Rolls (2004)).
Hence, we also make use of the visual search
paradigm to obtain insights into the observers’ rea-
soning processes while solving a visual task. We
Pallarés, V., Rami, L. and Dempere-Marco, L.
A Novel Method for Disentangling Strategies from Visual Search.
DOI: 10.5220/0005821602710276
In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 4: BIOSIGNALS, pages 271-276
ISBN: 978-989-758-170-0
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
271
argue that this approach opens up the possibility to
study the eye movements as a hallmark of cognitive
function. To this end, a new approach for knowledge
gathering from visual search is presented and evalu-
ated here, as well as applied to a simple visual search
task. The novelty of the proposed work lies on the
definition of a generalized feature domain in which
the dynamics of the eye movements patterns are an-
alyzed. This analysis includes both visual cues (fea-
tures) of the stimulus and spatial characteristics of the
patterns, which enrich the characterization of the eye
movements providing a new generalized feature do-
main in addition to the more common event-related
spatial domain (Holmqvist et al., 2011).
2 METHODS
2.1 Experimental Set-up
To assess the validity of the proposed method, we
have considered a double conjunction search task akin
to that presented by Hu et al. (2003). We have, how-
ever, modified such task in this study by systemat-
ically manipulating the image content in order to fa-
cilitate the emergence of optimal feature-based strate-
gies during visual search. Figure 1 illustrates several
test cards, together with the experimental protocol,
used in this study. The goal of the task is to find the
unique object (i.e. a square) that presents two specific
features (i.e. the colors blue and yellow) among a set
of distractors (i.e. the rest of squares).
Two groups of subjects have been considered in
this study: 7 young participants (2 male/ 5 female,
ages [22-38]), and 7 elderly subjects (2 male/ 5 fe-
male, ages [58-68]). Their eye movements have
been monitored with an eye-tracker (Tobii X120,
f
s
= 120 Hz) while performing the task. The sub-
jects were seated 65 cm in front of a 17-inch monitor
(screen resolution 1024 × 768 pixels) and a chin-rest
was used to prevent head movements.
2.2 Extending the Hot Spot Framework
to a Generalized Feature Domain
A conceptual framework was presented in (Hu et al.,
2003) based on the hypothesis that there is a direct
relation between visual attention and oculo-motor ac-
tion. In this approach, which received the name of
hot-spot framework, the fixation events from the scan-
path (originally defined in the spatial domain) are pro-
jected onto a new feature domain (in this particular
case, the color domain). The prevalence of such fea-
tures in the visual scan-path —defined as a sequence
of fixations and saccades— are determined, which
can unveil the existence of the underlying strategies
pursued by the observer.
Given a scan-path ξ = hx
i
, y
i
,t
i
i, where each fix-
ation is centered at hx
i
, y
i
i and with a dwell time t
i
,
it is possible to extract the prevalence of a particular
feature f
k
as
T ( f
k
) =
hx
i
,y
i
i∈ξ
hx
i
,y
i
i∈(x
i
,y
i
)
t
i
P
i
f
k
(x, y)
!
(1)
where (x
i
, y
i
) is the foveal field of the fixation cen-
tered at hx
i
, y
i
i considering all the pixels within a 2
visual angle, and P
i
f
k
is a probability distribution func-
tion.
Those features that are relevant, not only from a
bottom-up perspective but also in conjunction with
top-down aspects, can be extracted by defining, in the
feature domain F, a density function Γ. To do this,
the prevalence of a feature point f
k
must be normal-
ized by a factor that represents the absence of any
predetermined strategy. This normalization factor is
computed by considering the exploration of the whole
card with no over/under scanning of any part, and it is
calculated as
T
0
( f
k
) =
hx
i
,y
i
i∈Image
hx
i
,y
i
i∈(x
i
,y
i
)
P
i
f
k
(x, y)
!
(2)
25% 5%10%15%20% 5%25% 25% 25%
Figure 1: The test cards are composed of squares distributed in a 10 × 10 grid with a gap between adjacent squares. Each
of the squares in the stimulus card is defined by two out of four possible colors: red, green, blue and yellow. A sequence of
nine different cards is presented to the participants following the above protocol, and each card is displayed until the target is
found. As illustrated in the example, four out of the nine cards have an equal global amount of color each (i.e. 25% prevalence)
whereas in the other five cards, the amount of yellow diminishes gradually, from 25% to 5% prevalence in 5% steps. Such
decrease intends to elicit the emergence of color-based search strategies as a result of its increasing visual saliency.
BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing
272
T
0
describes, then, the prevalence of a feature in a
scan-path which shows a systematic exploration of
the test card with no preference for any features of
the stimulus. The density function is finally defined
as
Γ( f
k
) =
T ( f
k
)
k
T ( f
k
)
T
0
( f
k
)
k
T
0
( f
k
)
=
T
0
( f
k
)
T
0
0
( f
k
)
(3)
T
0
( f
k
) characterizes the prevalence of the feature f
k
in the observers’ scan-path, as Equation (1), but now
considering the specific properties of the stimulus.
All in all, by turning T ( f
k
) and T
0
( f
k
) into prob-
ability density functions, Γ can be explained in terms
of signal detection theory (Hu et al., 2003). In par-
ticular, T
0
( f
k
) can be interpreted as the detected sig-
nal, which also contains a noise signal T
0
0
( f
k
). The
likelihood function in our case corresponds to the ra-
tio between T
0
( f
k
) and T
0
0
( f
k
). Thus, Γ > 1 would
be indicative of certain feature preference in that the
likelihood of paying attention to a particular feature
would be higher than that expected by chance. In this
work, we extend the original formulation in order to
capture strategies developed on any feature domain,
which also includes spatial patterns present in the vi-
sual scan-paths. Moreover, our approach can be ap-
plied to any eye movement event, not only fixations.
For this particular task, two specific feature do-
mains have been analyzed. On one hand, the preva-
lence of color-based patterns is assessed in the color
domain, as in (Hu et al., 2003). This is particu-
larly relevant in our study given the decrease in the
amount of one of the two target colors, which entails
the emergence of an optimal strategy based this fea-
ture. On the other hand, and in contrast to the origi-
nal work, spatial-temporal patterns have also been ex-
tracted from the scan-path and are projected onto a
spatial feature domain. This has allowed us to reveal
and characterize the emergence of certain stereotypi-
cal behaviors such as reading patterns or column-wise
exploration.
2.2.1 Color Domain
The emergence of strategies in the feature domain
is assessed from Equation (1) by considering color
as the visual cue which defines such domain. In
this study, the probability function P
i
f
k
(x, y) is de-
fined such that for each fixation hx
i
, y
i
i, P
i
f
k
(x, y) = 1
(x, y) (x
i
, y
i
), and else P
i
f
k
(x, y) = 0. As previ-
ously stated, this implies that the relevance of each
color is proportional to the attention received (see
Equation (1)), while it also depends on the character-
istics of the scene (see Equation (2)). Therefore, the
density function of each color is defined as in Equa-
tion (3). For these particular stimuli 1 k 4, rep-
resenting the four colors (i.e. red, green, blue and
yellow).
2.2.2 Spatial Domain
Similarly, the emergence of spatial-temporal patterns
is explored in this study. It is worth noting that sac-
cade direction conveys information about the geomet-
rical structure of the scan-paths, which in turn, may
reflect the existence of underlying systematic behav-
iors. Moreover, the analysis based on the considera-
tion of spatial positions (i.e. the hx
i
, y
i
i fixation co-
ordinates) is not traslation invariant. By using sac-
cade direction instead, the outcome is less sensitive
to translation and thereby less dependent on the local
characteristics of the stimulus. Thus, in order to ex-
plore this scenario, the angles of the saccades present
in the scan-path are analyzed, which give rise to a
newly defined feature domain of saccade directions.
To project the saccade events of the scan-path onto
this feature domain, Equation (1) and Equation (2) are
recalled. The saccade direction is calculated with re-
spect to the horizontal. The angular range of 360
is discretized into 12 intervals spanning 30
and cen-
tered at 0
, 30
, 60
, 90
, ..., 330
. Each pair of op-
posed angles is considered as the same spatial direc-
tion, and defines one out of 6 possible directions of
the eye movements. Since the saccade time length is
substantially short and remains largely constant, t
i
is
taken as a constant. The prevalence of each saccade
direction in the scan-path is then described by
T ( f
k
) =
hsacc
i
i∈ξ
P
i
f
k
(4)
where sacc
i
are the saccades extracted from the scan-
path ξ, and P
i
f
k
is a probability distribution function,
which is defined such that P
i
f
k
= δ(d f
i
k
) (i.e. P
i
f
k
= 1
if the direction of saccade i is k, and else P
i
f
k
= 0).
This prevalence must be normalized again by a fac-
tor that takes into account all the possible directions
for the saccades in the scan-path. This operation will
weight the saccade direction f
k
according to its dom-
inance when considering a strategy-free scan-path.
Then, Equation (2) can be rewritten following an anal-
ogous reasoning to that described in previous sections
as
T
0
( f
k
) =
hsacc
i
i∈Image
P
i
f
k
(5)
In this occasion, the lack of strategy is modeled as a
homogenous probability of saccadic programming in
A Novel Method for Disentangling Strategies from Visual Search
273
Horizontal
Diagonal 1
Diagonal 2
Vertical
Diagonal 3
Diagonal 4
Red
Green
Blue
Yellow
A
time(s)
Γ(%)
0 1 2 3 4 5 6 7
0
25
50
75
100
Color Domain
0
25
50
75
100
0 1 2 3 4 5 6 7
Spatial Domain
time(s)
Γ(%)
B
time(s)
Γ(%)
0 5 10 15 20
0
25
50
75
100
Color Domain
0 5 10 15 20
0
25
50
75
100
Spatial Domain
time(s)
Γ(%)
C
0
1 2
3
4
5 6
7
8 9 10
0
25
50
75
100
time(s)
Γ(%)
Color Domain
0
25
50
75
100
1 2
3
4
5 6
7
8 9 10
Spatial Domain
time(s)
Γ(%)
D
0 2 4 6 8 10 12 14 16
0
25
50
75
100
time(s)
Γ(%)
Color Domain
0
25
50
75
100
0 2 4 6 8 10 12 14 16
Spatial Domain
time(s)
Γ(%)
25%
25%
25%25%
20%
27%
26.5%26.5%
5%
32%
31.5%31.5%
5%
32%
31.5%31.5%
Figure 2: Illustration of four scan-paths that have been analyzed by projecting the information from fixations and saccades
in the color and spatial domains, respectively. [Left] stimulus cards with the saccades trajectories superimposed; [center] the
density function corresponding to each color in the feature domain; [right] the density function corresponding to each saccade
direction for the 6 possible directions. A density value over 50% is accepted as an indicator of pursuing a strategy.
any of the 6 possible directions. By again turning,
T ( f
k
) and T
0
( f
k
) into probability density functions a
probability density function Γ akin to that in Equation
(3) has been defined.
3 RESULTS
The analysis of the eye-tracking data in both domains
(i.e. color and spatial feature domains) has allowed us
to identify a number of stereotypical strategies, which
are shown in Figure 2. Interestingly, the proposed
method permits us to investigate both dynamical as-
pects of the visual search strategies as well as the
aggregate behavior which accounts for the complete
scan-path. In the first case, the density function Γ(t)
is evaluated by considering all of the fixations from
the scan-path occurred up to time t. Thus, it is the
dynamical evolution of the function Γ(t) what is de-
picted in Figure 2. In order to normalize the curves
into a common scale, the density functions have been
divided by the number of features n, which conform
the feature domain at hand, i.e. n = 4 for color and
n = 6 for saccade directions.
Figure 2 shows several paradigmatic examples
which correspond to individual visual scan-paths over
four different test cards. The first aspect that must be
noted is the existence of a transient period, which is
usually followed by a stationary state. It is during this
stationary state that the strategy becomes stable and
can be readily identified. Out of the four cases illus-
trated, Figure 2.A (a card presenting a 25% of each
color), does not reveal the presence of any strategy,
i.e. neither color-based nor any spatial-temporal pat-
tern. In fact, after a short transient period, the density
function for all four colors converges to a 25%, thus
revealing the absence of any feature preference. Sim-
ilarly, in the spatial domain no feature preference has
been identified, which is in agreement with the visual
scan-path (see left panel) in which no clear strategy
can be identified.
In contrast, Figure 2.B shows another trial (20% of
yellow) in which a spatial-temporal pattern emerges
from the visual scan-path. In this case, a column-wise
reading pattern can be identified (see left panel). The
proposed method clearly reveals such a strategy (note
the overall dominance of the density function corre-
sponding to the vertical saccade direction). Notably,
BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing
274
25% 25% 25% 20% 15% 10% 5% 5% 25%
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
25% 25% 25% 20% 15% 10% 5% 5% 25%
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Color DomainSpatial Domain
Pr
Pr
A
B
25% 25% 25% 20% 15% 10% 5% 5% 25%
-20
0
20
40
60
80
100
Completion time (s)
Young
Elderly
Figure 3: (A) Probability estimates of developing spatial-based and color-based strategies; the emergence of a strategy is
considered whenever a 50% density value is exceeded for young (blue) and elderly (red) populations. (B) Task completion
times for young (blue) and elderly (red) subjects.
this method does not only allow to identify whether
a particular strategy occurs but also when it emerges
and how (or whether) it is related to other strategies.
Similarly, the trial in Figure 2.C, which presents a dif-
ferent proportion of colors, shows a color-based strat-
egy. In this case, the density function corresponding
to yellow grows rapidly revealing a preference for this
color whereas no spatial-temporal pattern can be iden-
tified. Finally, Figure 2.D shows another trial (also
5% of yellow), in which a combination of color-based
and spatial-based strategies emerges. The analysis in
the color domain reveals a clear prevalence of yellow
in the visual scan-path while, at the same time, a row-
wise scanning pattern (i.e. dominance of horizontal
direction saccades) becomes apparent in the spatial
domain. The observer covered a large area of the card
pursuing this spatial-temporal pattern, but with a ten-
dency to focus on the most relevant feature given the
task at hand. This is an example of an optimal strat-
egy in that the most informative areas are fixated upon
while following a systematic strategy which allows
the subject to easily keep track of the items which
have already been visited.
Finally, the eye-tracking data from both popula-
tions is considered to assess possible differences be-
cause of aging. For each trial, two coefficients have
been computed, one describing spatial-based strate-
gies (i.e. horizontal or vertical saccade direction dom-
inance) and the other color-based strategies (i.e. yel-
low dominance). To this end, the stationary state and
its accompanying strategy in each of these domains is
characterized by the averaged density function value
during the last 500 ms of the trial. In the case of the
spatial domain, the maximum between the horizon-
tal and vertical saccade density functions is consid-
ered. Whenever Γ > 50%, we consider that a strategy
has been actively pursued. Each trial of the exper-
iment has been considered as a trial of a Bernoulli
process interpreted as a success whenever Γ > 50%.
This has allowed us to evaluate for a particular strat-
egy the probability of being pursued by a population.
The results obtained (Figure 3.A) show some ev-
idence of notable differences among the two popula-
tions. In particular, the probability of developing a
spatial strategy is higher for the elderly population
for almost any kind of card, while for young par-
ticipants spatial-based strategies tend to decrease as
color-based strategies emerge. In short, the elderly
apparently undergo more systematic searches. Be-
yond visuo-motor aspects, and the large variability
observed in the data, such systematic patterns tend to
be accompanied by longer overall search times (as can
be seen in Figure 3.B).
4 CONCLUSIONS
In this work, we provide evidence supporting the no-
tion that it is possible to unveil, and mathematically
characterize, the strategies pursued by subjects while
solving complex visual tasks. A novel experimental
paradigm based on a double conjunction search, in
which one of the visual cues is systematically manip-
ulated, has been proposed. Both spatial-temporal and
feature-based patterns are studied by considering a
common methodological framework. The method has
been evaluated on two groups of subjects, who differ
in their age. The proposed method has allowed us
to: 1) characterize the strategies that are employed, 2)
study the emergence, temporal deployment and domi-
nance of such strategies, and 3) assess differences be-
tween the two cohorts.
We suggest that the use of eye-tracking technol-
ogy may provide important insights into aging re-
search. There is evidence that a decline in various
aspects of cognition accompanies the aging process.
For instance, older adults show a loss of processing
speed (Salthouse, 1996) and a decline in several as-
pects of executive function (working memory, task
switching, inhibitory function), and reasoning, as re-
A Novel Method for Disentangling Strategies from Visual Search
275
viewed in (Sperling et al., 2011). We, thus, hypothe-
size, and provide preliminary evidence, that the visual
search paradigm can be used to probe such basic cog-
nitive functions and their relation to normal aging and
age-related neurodegenerative pathologies.
ACKNOWLEDGEMENTS
The authors acknowledge funding from the re-
search project TIN2013-40630-R (Spanish Ministry
of Economy and Competitiveness)
REFERENCES
Beauchamp, M., Petit, L., Ellmore, T., Ingeholm, J., and
Haxby, J. (2001). A parametric fmri study of overt and
covert shifts of visuospatial attention. NeuroImage,
14(2):310–321.
Boccignone, G. and Ferraro, M. (2004). Modelling gaze
shift as a constrained random walk. Physica A: Statis-
tical Mechanics and its Applications, 331(1-2):207–
218.
Carrasco, M. (2011). Visual attention: The past 25 years.
Vision Research, 51(13):1484–1525.
Corbetta, M., Akbudak, E., Conturo, T., Snyder, A.,
Ollinger, J., Drury, H., Linenweber, M., Petersen, S.,
Raichle, M., Van Essen, D., and Shulman, G. (1998).
A common network of functional areas for attention
and eye movements. Neuron, 21(4):761–773.
Deco, G. and Rolls, E. (2004). A neurodynamical cortical
model of visual attention and invariant object recogni-
tion. Vision Research, 44(6):621–642.
Deco, G. and Zihl, J. (2006). The neurodynamics of visual
search. Visual Cognition, 14(4-8):1006–1024.
Dempere-Marco, L., Hu, X., and Yang, G.-Z. (2011). A
novel framework for the analysis of eye movements
during visual search for knowledge gathering. Cogni-
tive Computation, 3(1):206–222.
Deubel, H. and Schneider, W. (1996). Saccade target
selection and object recognition: Evidence for a
common attentional mechanism. Vision Research,
36(12):1827–1837.
Engbert, R. and Kliegl, R. (2003). Microsaccades uncover
the orientation of covert attention. Vision Research,
43(9):1035–1045.
Grosbras, M.-H., Laird, A., and Paus, T. (2005). Cortical
regions involved in eye movements, shifts of atten-
tion, and gaze perception. Human Brain Mapping,
25(1):140–154.
Hoffman, J. and Subramaniam, B. (1995). The role of visual
attention in saccadic eye movements. Perception &
Psychophysics, 57(6):787–795.
Holmqvist, K., Nystr
¨
om, M., Andersson, R., Dewhurst, R.,
Jarodzka, H., and van de Weijer, J. (2011). Eye Track-
ing: A comprehensive guide to methods and measures.
OUP Oxford.
Hu, X.-P., Dempere-Marco, L., and Yang, G.-Z. (2003). Hot
spot detection based on feature space representation of
visual search. IEEE Transactions on Medical Imag-
ing, 22(9):1152–1162.
Laubrock, J., Kliegl, R., Rolfs, M., and Engbert, R. (2010).
When do microsaccades follow spatial attention? At-
tention, Perception, and Psychophysics, 72(3):683–
694.
Rayner, K. (2009). Eye movements and attention in reading,
scene perception, and visual search. Q J Exp Psychol
(Hove), 62:1457–506.
Salthouse, T. (1996). The processing-speed theory of adult
age differences in cognition. Psychological Review,
103(3):403–428.
Sperling, R., Aisen, P., Beckett, L., Bennett, D., Craft, S.,
Fagan, A., Iwatsubo, T., Jack Jr., C., Kaye, J., Mon-
tine, T., Park, D., Reiman, E., Rowe, C., Siemers, E.,
Stern, Y., Yaffe, K., Carrillo, M., Thies, B., Morrison-
Bogorad, M., Wagster, M., and Phelps, C. (2011).
Toward defining the preclinical stages of alzheimer’s
disease: Recommendations from the national institute
on aging-alzheimer’s association workgroups on diag-
nostic guidelines for alzheimer’s disease. Alzheimer’s
and Dementia, 7(3):280–292.
Treisman, A. and Gelade, G. (1980). A feature-integration
theory of attention. Cognitive Psychology, 12(1):97–
136.
V
˜
o, M.-H. and Henderson, J. (2010). The time course of
initial scene processing for eye movement guidance in
natural scene search. Journal of Vision, 10(3):1–13.
Wolfe, J. (1998). Visual search. In Pashler, H., editor, Atten-
tion, Studies in cognition. University College London
Press, London, UK.
Yang, G.-Z., Dempere-Marco, L., Hu, X.-P., and Rowe, A.
(2002). Visual search: Psychophysical models and
practical applications. Image and Vision Computing,
20(4):291–305.
Yuval-Greenberg, S., Merriam, E., and Heeger, D. (2014).
Spontaneous microsaccades reflect shifts in covert
attention. Journal of Neuroscience, 34(41):13693–
13700.
BIOSIGNALS 2016 - 9th International Conference on Bio-inspired Systems and Signal Processing
276